Predictive and prognostic markers of recurrence after resection of primary or metastatic colorectal cancer.

Abstract

447 Background: Current biomarkers for colorectal cancer sub-classify tumors based on single mutations, such as KRAS; however, studies of single mutations belie the molecular complexity of colorectal cancer in which an average of 14 key genes per tumor is dysregulated. We hypothesize that colorectal cancer may be molecularly sub-classified based on an oncogenic pathway prediction model in which tumors are grouped based on patterns of oncogenic pathway dysregulation/expression. Methods: Affymetrix microarray data from 850 patients with primary colorectal cancer from publically available datasets were combined using Bayesian factor regression modeling normalization. The activity of 19 separate oncogenic pathways was predicted among tumors to generate patterns of pathway activity for each sample. Mixture modeling was applied to these samples to identify subgroups of tumors with unique patterns of pathway dysregulation. Validation of subclasses was performed on a dataset of 133 primary and metastatic colorectal cancer samples of patients undergoing curative surgical resection at our institution. Tumors were subgrouped according to our previous model and recurrence free survival was calculated. In vivo validation was performed by treating NOD/SCID mice bearing patient derived tumors with everolimus with changes in tumor size calculated between day 0 and day 21. Results: Mixture modeling resulted in 6 individual subgroups of colorectal cancer based on pathway dysregulation. Kaplan Meier curves revealed that patients in subclass 4 had the poorest prognosis while patients in subclass 6 had the best prognosis (p=0.05). Further, tumors in subclass 4 were generally enriched for high mTOR pathway activation and patient derived explants from subclass 4 with high predicted mTOR activity were found to be sensitive to the MTOR pathway inhibitor everolimus. Conclusions: Prediction of oncogenic signaling pathway activity is a powerful tool that may be used to molecularly sub-classify colorectal cancer into biologically relevant subgroups. These subgroups have prognostic and predictive implications for recurrence following surgical resection and responsiveness to targeted therapy.